This document discusses why many machine learning models do not make it into production. It identifies several key reasons: 1. ML quality and performance concerns - Models may not meet necessary standards for accuracy, reliability, etc. 2. Simple alternative solutions - A simpler, non-ML solution may adequately address the problem. 3. Lack of management support - Projects may not receive the funding or resources needed to fully develop models. It then provides recommendations in several areas to help more models make it into production, including investing in labeled data, defining MLOps processes, bringing together interdisciplinary teams, and facilitating learning across roles.